Mailman 3 January 2008 - NumPy-Discussion

Warning: PyUFunc_FromFuncAndData: types, funcs,
by Neal Becker Jan. 15, 2008

Jan. 15, 2008

AFAICT, the 'types' and 'funcs' arguments to PyUFunc_FromFuncAndData are not copied. There is one example in numpy book showing them pointing to static storage, but the doc of PyUFunc_FromFuncAndData fails to point this out.

1 0

A question about ctypes and numpy
by william ratcliff Jan. 15, 2008

Jan. 15, 2008

Hi! I have been having some difficulty understanding how to deal with multidimensional ndarrays in structs with ctypes. Namely, I want to the define numpy arrays in python and have them operated on in C, and have them updated on the python side. Here is a toy example: c code fragment: foo.c typedef struct { double X[4][4]; double Y[4][4]; } matrices; //there could be other elements in the struct between the arrays, after them, etc. int correct(matrices* pt){ int i=2,j=1; pt->Y[i][j]=5.0; } python: import numpy as N import ctypes xdata=N.zeros((4,4),dtype='float64') ydata=N.ones((4,4),dtype='float64') class fiduc(ctypes.Structure): _fields_=[("X",some_ctype), ("Y",some_ctype)] pt=fiduc() pt.X=xdata.ctypes.data pt.Y=ydata.ctypes.data foo = N.ctypeslib.load_library('foo.dll', '.') foo.correct(ctypes.byref(pt)) print xdata print ydata My question is: If I want to define a class in python and later to have my c code operate on an instance of it, then I need to know the c_type representation for the ndarray. I can't simply blast the data into the structure because due to memory alignment, structs in C are not guaranteed to have contiguous elements in memory--I also don't want to take a risk with for example my mingw generated dll having a different alignment for a structure than what a ctypes made using msvc expects. Thanks!!!!! Cheers, William

1 0

Nose testing for numpy
by Matthew Brett Jan. 14, 2008

Jan. 14, 2008

Hi, I've just finished moving the scipy tests over to nose. Thinking about it, it seems to me to be a good idea to do the same for numpy. The advantages of doing this now are that numpy and scipy would be in parallel, that we can continue to have one testing system for both, and that it would be clear to both numpy and scipy developers that they should not use NumpyTest but the nose test framework. At the moment, you can still find yourself using the numpy test framework in scipy, and the tests will work - but it should be deprecated. Ideally, to make a clean switch, I think using numpytest should raise an error. One issue is that the scipy nose test labels use decorators, for which the standard syntax requires python 2.4. To avoid this, we can just apply the decorators with the def test_func(): pass test_func = dec.slow(test_func) syntax. It's a rather moot point, as the decorators are mainly used to define test level, there's only one extra test found for numpy.test(10) compared to numpy.test(). I think I could do the port in quite a short time. Any thoughts? Matthew

3 4

Update on using scons to build numpy
by David Cournapeau Jan. 14, 2008

Jan. 14, 2008

Hi there, A quick email to give an update on my work to build numpy with scons. I've finished a few days ago to make my former work a separate package from numpy: it was more work than I expected because of bootstrapping issues, but I can now build numpy again with the new package on Linux. The separate package is called numscons, and is developed using launchpad for bug tracking and source code repository: https://code.launchpad.net/numpy.scons.support You can find tarballs for releases there (you want the last one, that is 0.2.2 at that time): https://launchpad.net/numpy.scons.support/+download And you need to use the branch build_with_scons to build numpy with numscons (the way to build numpy with scons is as before: you use setupscons.py instead of setup.py). I am now working on cleaning some things in the code (in particular, all the changes I had to do to scons itself are now being worked on to be integrated upstream, so that upstream scons will be used instead of our own cannibalized copy). cheers, David

2 2

subclassing matrix
by Basilisk96 Jan. 13, 2008

Jan. 13, 2008

Hello folks, In the course of a project that involved heavy use of geometry and linear algebra, I found it useful to create a Vector subclass of numpy.matrix (represented as a column vector in my case). I'd like to hear comments about my use of this "class promotion" statement in __new__: ret.__class__ = cls It seems to me that it is hackish to just change an instance's class on the fly, so perhaps someone could clue me in on a better practice. Here is my reason for doing this: Many applications of this code involve operations between instances of numpy.matrix and instances of Vector, such as applying a linear- operator matrix on a vector. If I omit that "class promotion" statement, then the results of such operations cannot be instantiated as Vector types: >>> from vector import Vector >>> import numpy >>> u = Vector('1 2 3') >>> A = numpy.matrix('2 0 0; 0 2 0; 0 0 2') >>> p = Vector(A * u) >>> p.__class__ <class 'numpy.core.defmatrix.matrix'> This is undesirable because the calculation result loses the custom Vector methods and attributes that I want to use. However, if I use that "class promotion" statement, the p.__class__ lookup returns what I want: >>> p.__class__ <class 'vector.Vector'> Is there a better way to achieve that? Here is the partial subclass code: #---------- vector.py import numpy as _N import math as _M #default tolerance for equality tests TOL_EQ = 1e-6 #default format for pretty-printing Vector instances FMT_VECTOR_DEFAULT = "%+.5f" class Vector(_N.matrix): """ 2D/3D vector class that supports numpy matrix operations and more. Examples: u = Vector([1,2,3]) v = Vector('3 4 5') w = Vector([1, 2]) """ def __new__(cls, data="0. 0. 0.", dtype=_N.float64): """ Subclass instance constructor. If data is not specified, a zero Vector is constructed. The constructor always returns a Vector instance. The instance gets a customizable Format attribute, which controls the printing precision. """ ret = super(Vector, cls).__new__(cls, data, dtype=dtype) #promote the instance to cls type. ret.__class__ = cls assert ret.size in (2, 3), 'Vector must have either two or three components' if ret.shape[0] == 1: ret = ret.T assert ret.shape == (ret.shape[0], 1), 'could not express Vector as a Mx1 matrix' if ret.shape[0] == 2: ret = _N.vstack((ret, 0.)) ret.Format = FMT_VECTOR_DEFAULT return ret def __str__(self): fmt = getattr(self, "Format", FMT_VECTOR_DEFAULT) fmt = ', '.join([fmt]*3) return ''.join(["(", fmt, ")"]) % (self.X, self.Y, self.Z) def __repr__(self): fmt = ', '.join(['%s']*3) return ''.join(["%s([", fmt, "])"]) % (self.__class__.__name__, self.X, self.Y, self.Z) #### the remaining methods are Vector-specific math operations, including the X,Y,Z properties... Cheers, -Basilisk96

4 8

Template system
by Charles R Harris Jan. 12, 2008

Jan. 12, 2008

Hi All, I'm thinking of adding the template system I pulled out of Django to Numpy to use in place of the current code generator. Its advantages are documentation, flexibility, and maintainance. The code totals about 470 KB, comes with a BSD license, and is compatible with Python >= 2.3. I want to bring this up on the list because I'm sure there will be objections along the lines of "why do we need that"? So I want to see how strenuous the objections are. Chuck

3 8

wrap Python/numpy to C/C++. mingw. elmer.
by mani sabri Jan. 12, 2008

Jan. 12, 2008

Hi Sorry for irrelevant subject. I found elmer when I was googling for something to wrap my python/numpy code to C/C++ automatically because I want a dll for some application and I know nothing about python Api (shame on me). Does anybody successfully compiled elmer on windows with mingw? Or is it anything better than elmer? (mingwISHER!/windowsISHER!) I will appreciate any suggestion. Regards Mani

2 1

example of ufunc c-code?
by Neal Becker Jan. 11, 2008

Jan. 11, 2008

Where can I find a (hopefully simple) example of c-code for a ufunc?

2 2

Setting WRITEABLE flag on array scalar
by Stefan van der Walt Jan. 10, 2008

Jan. 10, 2008

Hi all, We currently use an array scalar of value False as the mask in MaskedArray. I would like to make sure that the mask value cannot be modified, but when I try import numpy as np x = np.bool_(False) x.flags['WRITEABLE'] = False I am warned that you "Cannot set flags on array scalars.". Is there any reason why this is prohibited? Which is the best way around it? Thanks for any feedback. Stéfan

3 3

BzrMirror page on numpy wiki
by David Cournapeau Jan. 10, 2008

Jan. 10, 2008

Hi, I started a page on how to use bzr, with a link to an import I have done on the numpy trunk: http://scipy.org/scipy/numpy/wiki/BzrMirror I thought it would be useful for people not familiar with DVCS, and are curious to try. In perticular, since I have done the svn->bzr import, the distributed bzr binary will work out of the box on windows and max os X (leopard). For now, I only wrote how to do basic things like getting the code/log/annotation/status, nothing related to actual hacking of the code inside the numpy code. Also, the mirror is pure http, which means it will be pretty slow to get the code the first time (I don't have access to more than bare http). For information, a full checkout (with all the history since 2001, trunk only) takes 30 Mb, without counting the working tree. cheers, David

3 5